WSU Center for Mathematical Modeling

1
Real-Time Sensor Driven Non-Invasive Diagnostics
for Biomedical Application
Dr. Edward Griffor DaimlerChrysler and WSU
Computer Science and
Prof. Loren Schwiebert Department of Computer
Science Wayne State University
Dr. Maik Hüttemann Center for Genetics and
Molecular Medicine Wayne State University

• WSU Center for Mathematical Modeling
• in the Medical Sciences

2
Project Real-Time Sensor Driven Diagnostics for
Biomedical Applications

• The goal of this proposal is
• to develop a non-invasive automated sensor system
• measure organismal oxygen uptake and utilization,
  and
• adjust input oxygen according to demand using an
  animal model

3
Project Support

• DaimlerChrysler Corporation
• Wayne State University School of Medicine
• Applied for
• WSU Seed Funding 2006-7
• External Application Funding 2008-2011

4
The Test Case

• Animal model knockout mouse that lacks a
  respiratory system specific gene called CcO4-2
• Protein product of this gene is part of the
  cytochrome c oxidase complex, the enzyme which
  transfers energy equivalents from food to the
  oxygen we breathe
• Absence of the gene leads to decreased oxygen
  utilization in the lung hence
• Decreased aerobic energy metabolism in the lung
  appears to be caused by the absence of CcO4-2

5
Aim 1 Assess O2 Utilization in Knockout Mice

• Assess oxygen utilization in CcO4-2 knockout and
  wild-type mice using a modified plethysmograph

Y-axis Turnover (per second)
X-axis Cytochrome c (micro M)
6
Aim 2 Optimize Accuracy thru Sensor Fusion

• Integrate readings from multiple sensor inputs to
  achieve optimal blood O2 levels

Sensors for Experimental Setup
7
Aim 3a Model-Driven Approach to Sensor Fusion

• Develop a model-driven approach to sensor data
  fusion that can be applied to a wide range of
  problems
• Develop an architecture for model-based querying
  in sensor networks.

8
Aim 3b Mathematics of Constraints

• The key queries are range queries
• Using a probabilistic model, we compute the
  probability that a value lies within the
  specified range
• If this probability is very high, we are
  confident that the predicate (the value lies in
  the range is true)
• Similarly, if the probability is very low, we are
  confident that the predicate is false
• If we do not have enough information to answer
  this query with sufficient confidence then we
  need to acquire more data from the sensor network
• The probability can be computed in two steps
  First, we project the PDF to a density over only
  attribute Xi by calculating the total integral of
  Projection gives us the PDF over only Xi. We can
  then compute simply by

9
Summary

• O2 and metabolic sensing are the focus of these
  applications
• The sensor technology is generally available and
  relatively inexpensive
• Good framework for proving out the proposed
  information processing architecture
• This framework has several high priority
  applications in the life sciences, including the
  monitoring/detection of
• O2 regulation for pre-term births
• Lung Disease
• Other disease involving O2 alterations, such as
  neurodegenerative disease and cancer

10
Research Opportunities

• Studies related to lung pathophysiology and
  oxygen metabolism
• Sensor data processing from multiple sensing
  sources
• Robust data models toward multivariate regression